GRU and LSTM Based Adaptive Prediction Model of Crude Oil Prices: Post-Covid-19 and Russian Ukraine War

The crude oil prices, which were stable for consecutive years, have been on a roller coaster since COVID-19. Owing to supply chain crises caused by the pandemic, the war between Russia and Ukraine, and the mismatch between excessive monetary policies and environmental protection policies, oil prices fell into negative territory in early 2020 unprecedentedly and hit new highs in recent days. On account of its universal approximation ability for any nonlinear function, the neural network has received substantial attention in asset price prediction. As a data-driven model, there is no doubt that the neural network can digest the past to predict the future. However, it cannot effectively predict those distinctive patterns that did not appear before, which is the case right now. In order to address this problem, a grey box adaptive Recurrent Neural Network (RNN) model based on feedback control in the control engineering field is proposed in this paper to compensate for the prediction error of the neural network. According to the experimental data, the correlation coefficients of the Adaptive Long Short-Term Memory (ALSTM) and Adaptive Gate Recurrent Unit (AGRU) proposed in this paper are 0.9895 and 0.9886, respectively, and the Root Mean Square Errors (RMSE) of these two models are 3.2184 and 3.3546, respectively. Therefore, the proposed models can improve prediction accuracy.